Parametric dictionary design using genetic algorithm for biomedical image de-noising application

Abstract

Due to their potential to generate sparse approximation of signals, overcompelete representations have particularly became an interesting issue in signal processing theory. Choosing an appropriate dictionary is a crucial point in sparse approximation methods. During the previous processes it has been shown that an incoherent dictionary is suitable for this method but recently, using parametric dictionary has gained much more interest so a set of parametric functions has been used to design this kind of dictionary. This paper discusses the use of different characteristics of equiangular tight frame (l1 norm of each frame) as a new objective function to find the best parameter. Using these characteristics helps us to eliminate the iteration method which has been used in previous experiments. In order to reduce this problem, we use Genetic Algorithm (GA). By implementing this algorithm in Multi-scale Gabor Function (MGF), we can reach to better results in comparison with other studies in this case. We applied both the initial and our designed dictionary on a Biomedical Image case study and compared the results to show the advantages of using this method in biomedical cases. © 2011 Springer-Verlag.